Introduction

In the previous chapter I introduced my forward converter that will take the job of generating +/-18V (ish) supplies from the incoming 20V rail. This was a non-isolated design, but draws on the likes of ATX power supplies. The output of this supply is still rather noisy but that isn’t a big problem: the idea has always been to sub-regulate the raw output of the switching regulator to +/-15V using linear regulators to provide clean analog supplies to the various sense amplifiers in the computer.

Let’s design these linear regulators now!

I’m going to heavily rely on the wonderful pages of Elliott Sound Products, where he (I assume it’s actually Elliott himself) walks through linear regulator design in detail.

The basic regulator

Being maximalist, I’m going to go with a feedback regulator from the previously mentioned page. Here’s the initial design:

The circuit closely follow Elliott’s recommendations, except for Q5 and Q4, which provides temperature compensation for the circuit. The output changes very little over the -25 to 125C temperature range:

There is one problem though: it needs at least about 2V of drop to operate. Now, this is why that is a problem: at 2A output current, a 2V drop would dissipate 4W on the poor power transistor, Q1. Can we reduce that?

The reason for this drop is that Q1 (and Q2) operate in an emitter-follower configuration. That means that it’s emitter voltage can’t be higher than one Vbe drop from it’s base voltage. And it’s base voltage is generated from the input voltage. Do that for both Q1 and Q2, plus add some margin to be able to generate sufficient base current to supply the 2A output and you end up with this ~2V drop. We can verify this by ramping the input voltage and see at what point the regulation kicks in:

There are two solutions to this problem. The first one is to use an LDO design, that uses a common-emitter configuration for the pass transistor, but that topology is tricky to get right and depends a lot on the gain of the error amplifier.

The other idea is to use a higher voltage rail to generate the base voltage than what is hooked up to the collector of Q1. Something like this:

With this change, the drop needed on Q1 is very small:

This trick is normally not available for linear regulator designs, but my situation is different: I do have a higher voltage (the 20V input to the forward regulator) lying around. So let’s use that!

A negative variant

Making a negative variant of the same regulator is only a matter of flipping some components and swapping all PNPs to NPNs and vice versa:

This works the same way, the performance is very similar (needs a little more drop due to the PNP pass transistor (Q4), but otherwise all is nice. Here’s the problem though: I don’t have a negative 20V supply just in my back packet. So how will this work?

Let’s look back at our forward converter:

There’s an extra winding on the central transformer that I didn’t use for anything! By hooking it up in series with the negative voltage output winding (L4), we can get a more negative voltage from it. This is what it looks like (with both linear regulators on the output):

Here I’m showing only the output portion of the design, the input doesn’t change and the drawing would be completely unreadable.

I should note that the switching inductor (L7) for this extra supply is not coupled to the other two (L5 and L6). That means that this negative output is not regulated and it’s output voltage depends on the load – which is very light. Consequently it’s voltage grows rather large, around 70-80V. For this reason, I’ve change Q16 to a transistor that can operate up to 150V. At the same time, I’m thinking, maybe I should somehow prevent this from happening: these voltage levels start to sound dangerous.

At any rate, that can wait, the simulator will not electrocute me.

Fine tuning

After running some tests, I realized that the temperature dependency of the forward regulator is too high: to accommodate the full -25 to 125 temperature range, I need either too high of a voltage drop on the linear regulators or tightening up the temperature dependency. After some fuffing around, I realized that the problem lies in the local power supply voltage for the controller. I’ve fixed that by adding an emitter-follower stage (Q18) to cancel out the temperature dependency of the pass transistor:

I also had to fine-tune the feedback to get the forward regulator output closer to 16V. With all that, here’s what the first 20ms of life looks like for the combined controllers:

Protection

I haven’t done much so far for over-current protection on these supplies: neither the forward converter nor the linear regulators have anything on their output. This was intentional: they were always to act as one unit, so intermediate protection is pointless. We are at the point though, where the outputs are available and I should consider this aspect.

The idea would be to put my usual over-current detection circuit at the output of the linear regulators, but have them feed back to the control of the switching regulator. This way, under high current settings it’s the switcher that reduces the output voltage, so the pass-transistors in the linear regulators don’t experience extra stress.

This is a bit harder to do on the negative side as there needs to be a voltage level translation, but here’s how it would look like for the positive and negative supplies respectively:

Here’s how it looks like on the positive side:

In the combined circuit, Q7 would be part of the AND gate with all the other limiters on the forward converter side and R11 would be just a single pull-up for all such sources.

On the negative supply, I need an extra inverter stage to get the polarity of the signal right, here Q8 would participate in the AND gate and R12 would be the common pull-up:

I think I will add PTC-type resettable fuses as well, but they should essentially never trigger: it’s just a belt and suspenders attitude.

Conclusion

I think this is a nice point to stop: I have a decent power supply for +/-15V. The voltage drop on the linear pass transistors is about 1V, so even at 2A load current, they would dissipate “only” 2W each. That’s something, I should be able to deal with using heat sinks, or maybe even the PCB itself.

Next time I will wrap up the power-related discussion by designing a reset controller (it is related to power through the incorporation of the power-good signals) and hopefully discuss the clock generation circuit.